A semiautomatic algorithm for three-dimensional segmentation of the prostate on CT images using shape and local...

Area of Science:

• Medical Imaging
• Radiology
• Computational Anatomy

Background:

• Prostate segmentation in CT images is crucial for treatment planning but is challenging due to low soft-tissue contrast.
• Manual segmentation is time-consuming and suffers from significant inter-observer variability.

Purpose of the Study:

• To develop and evaluate a semi-automatic, 3D prostate segmentation algorithm using shape and texture analysis.
• To improve the accuracy and efficiency of prostate segmentation in CT images.

Main Methods:

• Developed a 3D segmentation algorithm incorporating a point distribution model for prostate shape variation and local texture analysis.
• Utilized user inputs for initial prostate localization and trained the algorithm on 23 CT images, testing on 10.
• Evaluated performance against manual segmentations from two experts using Dice Similarity Coefficient (DSC) and Mean Absolute Distance (MAD).

Main Results:

• The algorithm achieved an average DSC of 88 ± 2% and MAD of 1.9 ± 0.5 mm.
• These results are comparable to the inter-expert variability (91 ± 4% DSC and 1.3 ± 0.6 mm MAD).
• Demonstrated fast, robust, and accurate performance without requiring prior intra-patient information.

Conclusions:

• The proposed semi-automatic algorithm provides an accurate and efficient method for 3D prostate segmentation in CT images.
• This approach has the potential to streamline diagnostic and therapeutic procedures by reducing segmentation time and inter-observer variation.
• The method shows promise for clinical applications requiring precise prostate delineation.